# VirtualBacon: "What is Bittensor? 2026 Update with Co-Founder Jacob Steeves (Const)"

- Title: What is Bittensor? 2026 Update with Co-Founder Jacob Steeves (Const)
- Channel: VirtualBacon (crypto investing channel; host is the channel owner)
- Date: about April 2026 (a mirror lists it as posted roughly 154 days before mid-September 2026; the host references the Covenant-72B run, Jensen Huang's All-In Podcast remarks, and the first halving)
- Length: about 1 hour (listed as "1 hr")
- URL: https://www.youtube.com/watch?v=YmxqgTXfpDo
- Jacob's start offset: none given; Jacob's first answer at about 1:00
- Method: YouTube blocked from this VM. Auto-caption text read from a rendered copy of the public watch page. No audio, so timestamps are estimated from position in the text and marked "≈".
- Format: research transcript. Host paraphrased; Jacob's key lines quoted verbatim (caption spelling cleaned).
- Speakers: HOST = VirtualBacon; JACOB = Jacob Steeves.

## Outline with timestamps

≈0:00 HOST frames the episode: his audience knew Bittensor from the 2023–24 AI wave; today is the "latest update" and how the vision now has a clear roadmap.

≈1:00 HOST: brief intro and how you came to found it.

≈1:10 JACOB: degree in maths and CS; in the early 2010s "it was pretty obvious to people that were paying attention that they had invented a new thing with this deep learning." Hinton and LeCun "were totally inspired by nature, by the brain. And so was I." Worked for DARPA on neuromorphic chips "kind of like what Extropic is doing." First Bitcoins 2014–15: "this whole other side of computer science, which is much more political freedom-based and also computational." Core idea: "how can we use this new language of money to solve problems in AI." "how can we actually invent a new type of computing around money."

≈3:30 JACOB on money as an algorithm. "you can organize matter effectively with currency, and that's what economies are." Cites Hayek: economies aggregate and organise resources. "It's an optimization technology really is money." "I was an algorithm person." Founded the Opentensor Foundation, launched the network in 2021. Terms he tries: "monetary computing," "incentive computing," "the monetary computer." "I'm just kind of making things up as I go in terms of what were the words."

≈5:40 HOST: Bitcoin's real innovation is coordination that no country can take over; AI needs that layer. Asks for an overview of subnets, since Templar (subnet 3) just finished a training run.

≈7:00 JACOB: subnets are markets for digital things that have no market today: inference speed, storage, "the value of a gradient to a machine learning model." Thought experiment: if you tried to build a gradient market from scratch you would end up building something like Bitcoin, then Bittensor. Goal: "a hierarchical stack of markets" from low-level commodities up to "this ultimate commodity," "the market for intelligence," "the TAO molecule." Yuma Consensus is the abstract design; subnets are "instantiations of people applying a mechanism through our abstract tooling."

≈10:30 JACOB explains Templar in depth: internet-scale training is one of the oldest ideas in AI and crypto; sharded model on buckets; miners contribute gradient updates. The hard part is not communication but permissionless adversaries: miners who "submit gradients that just destroy the model" or "pretend to do work." Templar built "a direct market for the value of an update to a model given the current state of the model." Names SparseLoCo (from DiLoCo) and a new compression scheme; notes someone used SparseLoCo to train across MacBooks at home.

≈17:00 HOST: highlights that this is pre-training from scratch, the largest so far, comparable to a 2023 Llama; the "escape" from a top-down AGI world.

≈20:00 JACOB: "not only does everyone have the opportunity to contribute, but also everyone has the opportunity to own it." Points to Stability AI's revenue-threshold licence as a possible model for Templar. "if you contributed to that run, or if you purchased the token to fund the development of the model, you get to own that, and that's viscerally on chain." "It's not just about us collaborating in sort of communist equality way. It's like we can make this capitalistic." On Jensen Huang's "folding at home" comparison: Folding@home "doesn't have any monetary incentive"; "when you add the economic component to the game, you can really scale." Path to a trillion-parameter mixture-of-experts model.

≈24:00 HOST: other interesting subnets people have not seen?

≈24:20 JACOB runs through subnets 1 to 5: SN1 meta incentive games (optimising matrix multiplication, CUDA kernels); SN2 verifiable inference ("you can use that in court"); SN3 Templar; SN4 Targon, trusted execution, the "Targon virtual machine," paper with Intel, "the prices are really cheap"; SN5 predictive commodity, "better than Polymarket a lot of the time." Then honesty about the long tail: "there's 128, and they're not all great." "some of them are scams and rug pulls and everything that you see in crypto. You give people freedom, and like you get amazing things out of that, but you also get all of the usual suspects." Affine: "This is my baby I'm building": a market where miners fine-tune and merge each other's models against benchmarks; recently trained "a very good model... better than GPT-3 and that came out of just a market." Then the pitch line: "an individual with a subnet done right can become a trillion-dollar company." Vision of the stack: "a hierarchically connected, completely the-guts-blown-out AI company."

≈33:00 HOST: incentives and economics; what is dynamic TAO?

≈33:30 JACOB: after giving the platform away, who decides which subnets get TAO? "what we tried to build is basically like an oligarchy that was going to vote on stuff. And that's a very inefficient solution." Do you ask "Jim Cramer" what it is worth? No. "we applied ourself again and we built a market for the subnets themselves." Each subnet gets a token and a pool; measure how well each "constrain[s] TAO, pulling TAO in and not releasing it." Emission from the coinbase "same as Bitcoin," 0.5 TAO per 12-second block post-halving, flows to subnets that "attract and sequester the most amount of TAO." "we built this sort of like flywheel machine." Supply shock: inflation gets locked in pools. 21 million cap, no pre-mine, halvings; when Kyle Samani asked why 21 million: "it's an homage to Bitcoin and also because it's just really understandable." "it's important in markets that people understand them." Subnet tokens copy the same rules.

≈40:00 HOST: how do new emissions reach subnets?

≈40:20 JACOB: a subnet is a bipartite graph, validators over miners, "proof of stake, proof of work hybrid," scored on a curve, cycled like "a genetic experiment." At the top level almost nothing is on chain except price and liquidity, so the chain measures "inflow": "a moving average of the amount of TAO that is sequestered and not lost from the pool." "we built a commodities market of commodity markets." The meta-game: "which subnet can you build a subnet that can attract inflow and not lose it?"

≈45:00 JACOB on culture: "it's kind of our cult is like we are incentive masters." Building exploit-resistant systems is "an art." "you have to live with your chest open to getting destroyed by this infinitely deep web of sybil resistant, permissionless, anonymous people that could be working for North Korea for all you know." When it works, "you turn them away from hackers and you turn them into engineers." dTAO turned subnets outward: "everyone went to Twitter"; "a machine that naturally, just by the incentive system itself, like goes and does marketing for you." A new class appeared: traders, "the worst degens you could imagine to like hyper-intelligent traders." Corporation analogy: "what if every single team in Google had a token and a market? And there was no managers anymore." "a swarm hive mind swarm of degen traders that show up at your desk and ask you what you're doing." Fair warning: subnet builders are "going quite insane at how intense it is... Everything is just weird black hole singularity that you're just getting pulled through."

≈52:00 HOST: this is a very AI way of designing, like reverse prompting from the goal.

≈52:30 JACOB agrees: ML is "define the thing that we want, produce a liquid layer underneath it that can adapt, and then let the thing underneath adapt to the goal... You can just define that exact same structure with money." Asked "Is it AI?": "Yes, it is AI. But AI is actually just pointing at that thing, too, which is nature... AI discovered it first. But it's not AI. That's just the way things work in nature." With crypto you can point the loop at "the entire workforce of the world," all computers, software, even proving you went to a place. "AI deeper with money."

≈56:00 HOST: how do big investors get involved, since there is no VC money?

≈56:20 JACOB: "you just buy TAO." Alpha tokens carry extra risk and return; "all those people that want to take risk are just working for people that are just holding TAO." "I'm not shilling my bags here." Why it matters that anyone can enter: closed labs "have already been sold out to the insiders... It's like fiat AI. I like to say that." Web3's real invention: "the ownership layer for the web." "it's not open source. It's a completely different beast."

≈59:30 HOST: AI will be the biggest investment, yet you cannot get exposure except an OpenAI IPO at a trillion.

≈60:00 JACOB: without ownership people have no relationship to AI "other than just as a customer, as cattle that suckles off the teat of the OpenAI credits." "I'm very afraid of that." "a completely non-transparent centralized global succubus of humanity where we just basically enslave ourselves because we become irrelevant. And we don't even know how that happened or who pulls the strings." "most of us are just going to not have any place in that world... It should scare us and we should be looking for different ways of building symbiosis between humans and the machine."

≈62:00 HOST closes; where to find Bittensor. JACOB: Twitter; the Discord; mine subnets; "get your OpenClaw... to mine Bittensor subnets"; "the future of Bittensor is subnets that are being run by agents and mined by AI agents"; papers on bittensor.com.

## Jacob's stage-worthy lines (verbatim, auto-caption source)

- "you can organize matter effectively with currency, and that's what economies are." (≈3:40)
- "an individual with a subnet done right can become a trillion-dollar company." (≈31:00)
- "we built a commodities market of commodity markets." (≈43:00)
- "what if every single team in Google had a token and a market? And there was no managers anymore." (≈48:00)
- "AI discovered it first. But it's not AI. That's just the way things work in nature." (≈53:30)
- "It's like fiat AI." (≈58:00)
- "as cattle that suckles off the teat of the OpenAI credits." (≈60:20)
- "we just basically enslave ourselves because we become irrelevant. And we don't even know how that happened or who pulls the strings." (≈60:50)
- "you have to live with your chest open to getting destroyed by this infinitely deep web of... anonymous people that could be working for North Korea for all you know." (≈45:40)